A major problem in developing effective treatments for mental illnesses is that specific drug effects are often obscured by the large degree of outcome variability due to placebo effects of treatment. Additionally, there is growing recognition of the therapeutic benefit of utilizing placebo effects in treating illnesses. In clinical practice prior to treatment, knowing the likelihood that a patient would benefit from nonspecific (i.e., placebo) effects can have an impact on treatment decisions. Also, knowledge of the amount of improvement during acute treatment that is due to non-specific effects would inform maintenance strategies. Consequently, the development of statistical methods that can distinguish specific and nonspecific effects of treatment will be important. Continuous advances in technology allow the development of sophisticated methodology for characterizing individuals with various psychiatric conditions; for example, brain imaging techniques provide high-resolution pictures of the structural and functional brain architecture. These complex high dimensional biological data, in conjunction with the modern development of flexible statistical methods that can accommodate such high dimensional data, presents an opportunity to obtain clinically useful characterization of patients experiencing placebo effects and to discover biosignatures for placebo response. Previously the investigators have developed methods for identifying placebo responders among drug treated subjects. Most are based on clustering and partitioning of trajectories of symptom severity during treatment. Although the developments could incorporate simple baseline covariates, the existing methodology is inadequate to deal with very high dimensional biological data such as brain images. The primary purpose of this application is to build on this foundation by developing approaches to increase the predictive power of baseline covariates that distinguish placebo response from specific response in drug treated subjects. The ultimate goal is to determine biosignatures of placebo response which we will define as patients' measures, linear combinations of such measures, or smooth, nonparametric functions of the measures, that differentially predict placebo and specific drug response. The aims are to develop models, computational methods for implementation, and analytic strategies for discovering such biosignatures, applicable to modern biomedical high-dimensional data. Our involvement in the EMBARC study (NIH-funded randomized placebo controlled clinical trial with PIs Trivedi, Weissman, McGrath, Parsey and Fava) provides access to an incredibly rich data source, consisting of extensive baseline measurements made on each subject (n=400). These data will allow for the development and testing of our methodologies for discovering biosignatures for placebo response using high dimensional data. Once developed and tested, they will be made available for research on other diseases and for different treatment modalities, such as psychotherapy. The new methods will facilitate efficient exploration of data that are typically available from standard randomized clinical trials.